AI Image Upscaler
Upscaling Workspace
Increase image resolution with browser-based ESRGAN super resolution.
Source Preview
AI Settings
Lower scale factors generally produce more reliable detail recovery.
Controls AI patch size. Padding remains enabled to reduce seam artifacts.
AI super resolution can generate plausible texture rather than recover information that was never present in the original. Very large images, higher scales and the Quality model can require substantial GPU memory and processing time.
Upscaled Result
Before / After Comparison
Export Result
PNG preserves transparency. JPEG always uses an opaque background.
About
The XAVERT AI Image Upscaler increases image resolution with ESRGAN neural networks running in your browser.
Main Features
- Upscale JPG, PNG and WebP images by 2×, 3× or 4×.
- Choose ESRGAN Slim for browser speed or ESRGAN Medium for a stronger quality/latency balance.
- Process images in padded patches to reduce blocking and visible patch seams.
- Use adaptive device-aware output pixel limits to reduce browser memory failures.
- Cancel active AI inference without reloading the page.
- Compare the original and upscaled result with an interactive slider.
- Preserve PNG/WebP transparency by resampling the source alpha channel separately when transparency is detected.
- Export the result as PNG, JPEG or WebP.
- Copy the final result as PNG when clipboard image writing is supported.
Privacy First
Your selected image is processed locally in your browser and is not uploaded to XAVERT. TensorFlow.js, UpscalerJS, the selected ESRGAN model configuration and its model weights are downloaded from third-party CDN infrastructure when needed. Browsers may cache these resources for later use.
AI upscaling is generative enhancement. It can create plausible details and textures, so the result should not be treated as a forensic reconstruction of missing source information.